Loading video...

Video Failed to Load

There was a problem loading this video. This could be due to a temporary network issue or the video might be unavailable.

A curious property of diffusion models: when a sample is mostly noise, the model’s best guess is close to the mean. So early on samples can be pulled toward the center of the dataset, even when no data exists there, before rebounding toward the data manifold.

Alec Helbling

11,720 subscribers

31,569 views • 7 days ago •via X (Twitter)

Anya Rossi• Live Now

Private livecam show

0 Comments

No comments available

Comments from the original post will appear here

Related Videos

The latent space of earlier generative models like GANS can linearly encode concepts of the data. What if the data was model weights? We present weights2weights, a subspace in diffusion weights that behaves as an interpretable latent space over customized diffusion models.

The latent space of earlier generative models like GANS can linearly encode concepts of the data. What if the data was model weights? We present weights2weights, a subspace in diffusion weights that behaves as an interpretable latent space over customized diffusion models.

Amil Dravid

94,296 views • 2 years ago

When asked if CANADA is in a recession, CARNEY says the data is just uneven What does this even mean 🇨🇦

When asked if CANADA is in a recession, CARNEY says the data is just uneven What does this even mean 🇨🇦

Melissa 🇨🇦

52,958 views • 3 months ago

This is what it sounds like living next to a data center. The video below was recorded at midnight, and the data center is situated next to 100s of residential homes.

This is what it sounds like living next to a data center. The video below was recorded at midnight, and the data center is situated next to 100s of residential homes.

Merissa Hansen

7,129,209 views • 4 months ago

In the data manifold, the shortest path between two points is a geodesic that pass through high density regions of data. Just like mass curves the space geometry, data also curves the space. In the example below we can see a geodesic being optimized between two points. 1/3

In the data manifold, the shortest path between two points is a geodesic that pass through high density regions of data. Just like mass curves the space geometry, data also curves the space. In the example below we can see a geodesic being optimized between two points. 1/3

Christian S. Perone

260,326 views • 1 year ago

Google is making progress on their diffusion models... It's now as good as a Gemini 2.0 Flash Lite. The writing is on the wall, a majority of language AI use in the future will be diffusion models.

Google is making progress on their diffusion models... It's now as good as a Gemini 2.0 Flash Lite. The writing is on the wall, a majority of language AI use in the future will be diffusion models.

Carlos E. Perez

193,952 views • 8 months ago

There's a truck driving around the RNC that emits the noise of a Data Center. Donald Trump said people who oppose Data Centers are "backwards and poor." Let's see how he likes having a Data Center right next door.

There's a truck driving around the RNC that emits the noise of a Data Center. Donald Trump said people who oppose Data Centers are "backwards and poor." Let's see how he likes having a Data Center right next door.

Texas Democratic Party

18,421 views • 5 days ago

There is NO DEBATE The Smith Machine is the best piece of equipment to perform Split Squats with The next best option…. ….is NOT EVEN CLOSE

There is NO DEBATE The Smith Machine is the best piece of equipment to perform Split Squats with The next best option…. ….is NOT EVEN CLOSE

Dean Turner

103,470 views • 6 months ago

🚨 Brad Garlinghouse, CEO of Ripple, stated that there is a 75% likelihood that by the end of April the company will be fully back on track and very close to final sign-off. Momentum toward regulatory clarity is building. $RLUSD | #XRP

🚨 Brad Garlinghouse, CEO of Ripple, stated that there is a 75% likelihood that by the end of April the company will be fully back on track and very close to final sign-off. Momentum toward regulatory clarity is building. $RLUSD | #XRP

CryptoSensei

31,584 views • 7 months ago

In flow matching, a coupling determines how noise and data samples are paired during training. The choice of coupling is important because it influences the geometry of trajectories at inference time. The simplest choice is the independent coupling, where noise and data points are paired arbitrarily. This can lead to curved trajectories as the model averages over many conflicting pairings. However, if we use optimal transport on batches of pairs, this leads to fewer ambiguous intersections that the model must resolve, leading to straighter trajectories at inference time.

In flow matching, a coupling determines how noise and data samples are paired during training. The choice of coupling is important because it influences the geometry of trajectories at inference time. The simplest choice is the independent coupling, where noise and data points are paired arbitrarily. This can lead to curved trajectories as the model averages over many conflicting pairings. However, if we use optimal transport on batches of pairs, this leads to fewer ambiguous intersections that the model must resolve, leading to straighter trajectories at inference time.

Alec Helbling

65,484 views • 4 months ago

When the game is on the line, LaMelo is one of the best defenders when he’s on an island…This should of been a no call 😮‍💨

When the game is on the line, LaMelo is one of the best defenders when he’s on an island…This should of been a no call 😮‍💨

Kyle Ingram

2,543,959 views • 1 year ago

Langevin Monte Carlo allows you to draw samples from a probability distribution using its log gradient ∇ log p(x). By performing a sort of gradient ascent with noise you can navigate around the distribution. Langevin MC is heavily related to modern diffusion models.

Langevin Monte Carlo allows you to draw samples from a probability distribution using its log gradient ∇ log p(x). By performing a sort of gradient ascent with noise you can navigate around the distribution. Langevin MC is heavily related to modern diffusion models.

Alec Helbling

96,981 views • 1 year ago

ALERT: City councilwoman cheered for after she walked out of a meeting to block a vote on a data center proposal in Kentucky. Melinda Wofford walked out of the Burgin City Council meeting, causing the council to lose its legal minimum of voters and ending the meeting before the data center vote could take place. The city attorney drafted an ordinance to annex several parcels of land for the Panattoni Data Centers, which the council had already voted against once. Residents are concerned about the data center's potential impact on air quality, noise, water, electricity demand, and local infrastructure. Wofford said she walked out intentionally and, "You can’t do business without a quorum.”

ALERT: City councilwoman cheered for after she walked out of a meeting to block a vote on a data center proposal in Kentucky. Melinda Wofford walked out of the Burgin City Council meeting, causing the council to lose its legal minimum of voters and ending the meeting before the data center vote could take place. The city attorney drafted an ordinance to annex several parcels of land for the Panattoni Data Centers, which the council had already voted against once. Residents are concerned about the data center's potential impact on air quality, noise, water, electricity demand, and local infrastructure. Wofford said she walked out intentionally and, "You can’t do business without a quorum.”

E X X ➠A L E R T S

51,062 views • 5 days ago

"The future of the planet will be decided by the people who own the data" This is why Twetch is so important, a thread 🧵

"The future of the planet will be decided by the people who own the data" This is why Twetch is so important, a thread 🧵

Twetch

12,523 views • 3 years ago

A fast-growing wildfire spread from the Sierra foothills toward the center of Reno early Sunday, as approximately 90,000 residents were urged to evacuate. As of Sunday, the Hawk Fire was 0% contained. The fire is believed to have been caused by a human, but the investigation is ongoing, authorities said.

A fast-growing wildfire spread from the Sierra foothills toward the center of Reno early Sunday, as approximately 90,000 residents were urged to evacuate. As of Sunday, the Hawk Fire was 0% contained. The fire is believed to have been caused by a human, but the investigation is ongoing, authorities said.

CBS News

41,381 views • 23 days ago

Here's a fast preview decoder for MiniMax-H3. It converts the model’s hidden image data into a rough preview while the video is generating, so you can quickly see the progress without waiting for the final result.

Here's a fast preview decoder for MiniMax-H3. It converts the model’s hidden image data into a rough preview while the video is generating, so you can quickly see the progress without waiting for the final result.

AI Search

32,201 views • 1 month ago

This is the noise currently coming from the 30MW data center in Dowagiac, Michigan. Now imagine what it will sound like at 340MW. 😱 Research indicates that noise from a 340MW facility could be heard by residents within a 2.5-mile radius—and certain low-frequency sounds from data centers can travel up to five miles from the source.

This is the noise currently coming from the 30MW data center in Dowagiac, Michigan. Now imagine what it will sound like at 340MW. 😱 Research indicates that noise from a 340MW facility could be heard by residents within a 2.5-mile radius—and certain low-frequency sounds from data centers can travel up to five miles from the source.

Kelly McCarty

402,070 views • 6 months ago

$Congrats to Mario 👏 for the accept of his ICLR paper on "Diffusion Graph Nets", predicting complex distributions of flow states on unstructured meshes Highly efficient, works even if the training data has only a fraction of the flow statistics.$

Congrats to Mario 👏 for the accept of his ICLR paper on "Diffusion Graph Nets", predicting complex distributions of flow states on unstructured meshes Highly efficient, works even if the training data has only a fraction of the flow statistics.

Thuerey Group at TUM

24,996 views • 1 year ago

This is literally insane So many people showed up to oppose a $6 billion dollar data center in Missouri they had to use bleachers The whole crowd yells and chants they don’t want the data center Festus City Council voted to approve the data center anyway right in their faces

This is literally insane So many people showed up to oppose a $6 billion dollar data center in Missouri they had to use bleachers The whole crowd yells and chants they don’t want the data center Festus City Council voted to approve the data center anyway right in their faces

Wall Street Apes

2,894,189 views • 5 months ago